Indium Phosphide
The bottleneck in AI infrastructure is spreading from inside the chips to the outside. Beyond GPUs and HBM, a more hidden shortcoming has emerged—Indium Phosphide (InP).
Executives at optical chip company Lumentum publicly stated at the RAISE Summit that the current delivery gap for Indium Phosphide (InP) laser chips has expanded to over 30%, with the shortage exceeding that of memory chips during the same period.
The competition for upstream capacity in the capital market has reached a fever pitch: NVIDIA has injected $2 billion each into Lumentum and Coherent for capacity expansion and locking in production; Nokia signed an agreement in August 2026 to acquire NXP's wafer fab in Arizona, USA, converting it into a dedicated Indium Phosphide production line.
As a III-V compound semiconductor that has been applied for many years, why has Indium Phosphide suddenly become an indispensable "choke point" material for AI computing networks? In the face of this supply crisis, how is the global industry chain responding?
I. The Rigid Demand for Computing Power in the Optical Interconnect Era
The physical laws of optical fiber communication dictate the irreplaceability of Indium Phosphide. The two spectral windows with the lowest transmission attenuation in quartz optical fibers are located at the 1310 nm and 1550 nm bands, respectively. The epitaxial wafers grown on an Indium Phosphide substrate have an electronic energy bandgap that precisely corresponds to this photoelectric conversion window. In the manufacturing of high-speed Electro-absorption Modulated Lasers (EML) with a single-channel rate reaching 200G, Indium Phosphide is currently the only substrate material with commercial feasibility.
The evolution of data center network speeds from 800G to 1.6T has brought about a nonlinear increase in material consumption. 800G optical modules are typically equipped with 4 to 8 Indium Phosphide EML chips; after upgrading to 1.6T, the number of channels expands to 16, and there is a comprehensive shift to high-power EML chips with a single-channel rate of 200G. The increased edge defect rate during the high-power wafer dicing and polishing process results in the physical loss of Indium Phosphide substrates and epitaxial wafers required to produce a single 1.6T optical module reaching 2.7 times that of 800G modules.
Under the Co-Packaged Optics (CPO) architecture, although silicon photonics technology has achieved the co-packaging of silicon photonic wafers and computing chips, since silicon is an indirect bandgap semiconductor and cannot emit light itself, an External Laser Source (ELS) must be introduced. Currently, all lasers capable of stably outputting high-power continuous light rely entirely on Indium Phosphide substrates. The introduction of the CPO route has increased the demand for external Indium Phosphide laser chips per computing server, thereby expanding the actual physical consumption of substrates at the source.
II. Triple Constraints Locking Down Capacity
In stark contrast to the demand pull, the upstream supply side faces multiple physical and industrial barriers.
In terms of market landscape and generational specifications, the prospectus of Beijing Tongmei shows that Sumitomo Electric Industries of Japan, JX Metals of Japan, and AXT of the US account for 91% of the global high-end optical communication-grade substrate market share. Overseas manufacturers are comprehensively advancing the transition to 6-inch production lines; however, according to announcements from listed companies such as Yunnan Germanium and Xingye Technology, domestic substrates are still dominated by 2-inch and 3-inch sizes. The 4-inch and 6-inch production lines are in the R&D and customer validation stages, and the control of crystal dislocation defect density still needs to be tackled.
Regarding capacity expansion and certification cycles, referencing the announcement of XianDao Microelectronics, the first-phase investment for a standardized 6-inch substrate production line usually exceeds 1 billion RMB. From substrate to epitaxial chips, thousands of hours of optical attenuation and aging tests are required. The certification cycle for downstream optical chip manufacturers for new substrate suppliers reaches 3 to 5 years, making it difficult to realize new capacity in the short term.
In terms of raw material supply, metal indium is a by-product associated with zinc ore smelting, lacking independent mines, and the purification threshold for 7N-grade high-purity indium is extremely high. As the spot price of upstream refined indium continues to rise, cost pressure is transmitted downwards in sequence, driving a significant increase in the quotes for mid-stream Indium Phosphide substrates.
III. Global Players Betting Heavily: Factory Acquisitions, Capital Injections, and Upgrades
Faced with a hard-to-fill supply gap, the global semiconductor industry chain is abandoning the conventional step-by-step capacity expansion model, turning to forced capacity locking through means such as factory renovation, large-scale capital allocation, and cross-industry factory acquisitions.
01. The Capacity Race Among Optical Chip Manufacturers
Overseas device and wafer foundry manufacturers have focused their competition on the generational transition to 6-inch production lines. Coherent chose to build a new 6-inch Indium Phosphide wafer fab in Texas, USA, with NVIDIA CEO Jensen Huang attending its groundbreaking ceremony.
Coherent is simultaneously upgrading its existing Indium Phosphide wafer production lines to advance the 6-inch process upgrade. In the logic of wafer processing, compared to the old 3-inch production lines, 6-inch Indium Phosphide wafers can yield nearly 4 times the number of chips per wafer, and the comprehensive manufacturing cost can be reduced by up to 50%. Coherent explicitly stated in its financial report that its goal is to double its internal Indium Phosphide capacity again by the end of 2027.
Another optical chip foundry giant, Lumentum, has avoided the long cycle of building a brand-new factory, choosing instead to directly acquire Qorvo's Gallium Arsenide (GaAs) wafer fab site in North Carolina, USA.
This Qorvo site was originally used for producing RF chips. Lumentum converted it into a dedicated Indium Phosphide production line supporting 4-inch and 6-inch processes by renovating its existing cleanrooms and process equipment. According to Lumentum's plan, the site will commence production in 2028, thereby filling the current dilemma of a delivery gap of over 30% for single-module EML chips.
Telecom system integrators have also begun to penetrate backwards into the wafer manufacturing segment. On August 5, 2026, local time, Nokia announced the signing of a definitive agreement to acquire NXP Semiconductors' semiconductor manufacturing campus in Chandler, Arizona, USA, planning to convert it into an Indium Phosphide optical chip production line. Nokia will start leasing part of the factory buildings for initial mass production in early 2027, and the overall acquisition closing is expected to be completed in the first quarter of 2029.
02. The Hundred-Billion-Yen Bet by Japanese Companies
At the upstream substrate manufacturing end, the Japanese giants holding a monopoly have initiated large-scale capital expenditures. In June 2026, JX Metals of Japan issued an announcement, declaring the allocation of 80 billion yen in idle funds generated from previous bond issuances, and planning to make an additional investment of 120 billion yen (approximately $749 million) before fiscal year 2030 to expand the domestic Indium Phosphide substrate capacity in Japan. Its established goal is to increase the total capacity to 7 to 10 times that of fiscal year 2025.
Sumitomo Electric Industries of Japan also publicly disclosed in July 2026 an additional investment of up to 18 billion yen. Sumitomo Electric abandoned the time-consuming plan to build a new factory, choosing instead to directly renovate its existing factory buildings and equipment at its Itami Works in Hyogo Prefecture, Japan, with the goal of increasing substrate capacity to 3.1 times that of fiscal year 2024 before fiscal year 2028.
03. Domestic Players Entering the Arena
Facing the barriers established by overseas giants through capacity and generational advantages, domestic enterprises with physical capacity and R&D foundations are making substantial inroads from four key nodes: raw material purification, substrate crystal growth, optical chip preparation, and IDM foundry.
In the upstream resources and substrate segment, #YunnanGermanium started implementing the "High-Quality Indium Phosphide Single Crystal Wafer Construction Project" in April 2026. The planned construction period is 18 months, expanding the production line to an annual output of 300,000 pieces (equivalent to 4-inch, including 6,000 pieces of 6-inch), ultimately reaching an annual output of 450,000 pieces. The plan for 2026 is to produce 180,000 pieces of Indium Phosphide wafers.
#Macrosilicon announced on August 5 that it has established a wholly-owned subsidiary, Zhuhai Mingyao, as the implementing entity for the Indium Phosphide semiconductor material project, with a registered capital of 10 million RMB. The project is divided into two stages: "preliminary R&D experiments" and "subsequent industrialization promotion." The first stage plans to invest 12 million RMB to build an R&D laboratory.
Xingye Technology announced in July that its wholly-owned subsidiary, Jiangsu Xingguang, acquired the Indium Phosphide substrate manufacturing and sales business of Qingdao Lidian for 55 million RMB.
#Piotech's subsidiary Guangdong Xianrui officially landed the annual production project of 40 tons of Indium Phosphide crystals on July 21, 2026, and completed the environmental impact assessment publicity. The project focuses on the mass production of core Indium Phosphide crystal raw materials, precisely filling the domestic upstream capacity gap for high-purity crystal materials.
On June 24, the 6-inch Indium Phosphide single crystal industrialization project of Tianjin Kuangjing Semiconductor was officially signed and landed. The base delivery was completed in July, and construction is steadily advancing. The project focuses on high-end 6-inch Indium Phosphide single crystal substrates with large sizes and high consistency.
In the optical chip manufacturing segment, the Indium Phosphide-based single-channel 200G EML laser chip developed by Yuanjie Semiconductor for 1.6T optical module demand is currently advancing pre-mass production preparations.
The generational evolution of semiconductor materials is constrained by inherent physical laws and engineering validation cycles. From NVIDIA's $2 billion capital injection to the hundred-billion-yen bets by Japanese material manufacturers, the essence of the tight supply and demand for Indium Phosphide is the chain reaction triggered by the physical light-emitting defects of silicon photonic materials during the evolution of AI computing networks towards 1.6T and CPO.
As the new capacity release nodes of overseas leading manufacturers are generally set between 2027 and 2030, the shortage pattern of Indium Phosphide substrates and high-end laser chips will continue in the capacity transition period over the next few years.
For the domestic industry chain, whether it can achieve physical breakthroughs in 4-inch yield improvement and 6-inch large-size crystal growth is the key to determining whether it can enter the global core computing power supply chain.
#IndiumPhosphide #Nokia #SemiconductorMaterials